---
license: cc0-1.0
language:
  - fa
task_categories:
  - text-to-speech
tags:
  - persian
  - farsi
  - audio8
  - codec-codes
pretty_name: Mana-TTS + GPTInformal as Audio8 codec codes
---

# Persian speech as Audio8 codec codes

Pre-encoded training data for [Audio8-Persian-TTS](https://github.com/ard9/Audio8-Persian-TTS):
each `shards/*.tar` holds one part of a source dataset, converted to Audio8 TTS codec indices
(`[10, T]` int16 arrays, ~21.5 frames per second) plus a JSONL manifest with the normalized text.
Using these files skips downloading and encoding the raw audio.

Sources (both CC0-1.0):
- [Mana-TTS](https://huggingface.co/datasets/MahtaFetrat/Mana-TTS) — Qharabagh et al., NAACL 2025
- [GPTInformal-Persian](https://huggingface.co/datasets/MahtaFetrat/GPTInformal-Persian)

The codes can be decoded back to audio with the Audio8 codec, so the original speakers' voices are
contained here. As requested by the dataset authors, do not use this data to imitate the speakers'
voices for malicious purposes.